Andalusia-MSMTU is an open-source dataset of multi-spectral time series from Andalusia (Spain) for spectral unmixing of Land Use and Land Cover (LULC) classes ready to train machine learning models. It was built based on the seven spectral bands of the MODIS sensors at 460m resolution from year 2013 (12 observations, one per month, for each MODIS pixel). Besides, every pixel includes a set of topographic and climatic metadata to further improve the spectral unmixing performance. Finally, we provide annotations at two hierarchical levels of LULC based on SIPNA LULC product. Level 1 includes 4 LULC classes: Artificial, agricultural lands, terrestrial wildlands and wetlands. Level 2 includes 10 LULC classes: Artificial, annual croplands, greenhouses, woody croplands, combinations of croplands and vegetation, grasslands, shrublands, forests, barelands, wetlands. The files provided in the dataset are: class_hierarchy.png: Hierarchical structure of the SIPNA-based LULC classes. The blue boxes represent the Level 1 (L1) classes. The green boxes represents the Level 2 (L2) classes. MODIS_grid.zip: Shapefile with the MODIS grid of Andalusia. Each pixel/square in the grid has associated a "square_id" as an identifier of the corresponding area. ts_files.zip: JSON files with the multi-spectral time series data. Each file corresponds to the multi-spectral time series of one pixel. The name of each file is the "square_id" of the corresponding pixel of the MODIS grid. Note that the file name is padded with 0 to the left up to 7 digits. metadata.zip: Metadata CSV file with the metadata information for each pixel. The "square_id" values are the identifiers of the pixels and the rest of the columns correspond to the metadata information collected. labels.zip: Annotations for level 1 and level 2 of the classification hierarchy. For each level we also provide the train, validation and test partitions used in our study.